MRI researchers interested in methods for portable imaging

A small rotating permanent magnet array with an inhomogeneous field is used to create a portable MRI scanner. The field variation of the inhomogeneous magnet is leveraged as the encoding field in the Y-Z plane instead of using additional gradient coils. This eliminates cooling requirements, and vastly reduces the weight and power requirements of the scanner.

The previously described 45 kg (36 cm dia. 35.6 cm length) Halbach cylinder magnet [1,2] (Fig. 1) uses 20 rungs of NdFeB magnets to create a 77.3 mT field (3.29 MHz proton freq.) with 32 kHz variation in the center slice 16 cm dia. FOV (mapped using small field probes). As the magnet is physically rotated around the object using a stepper motor, a RARE-type spin echo sequence (40 kHz, 256pt readout) records the generalized projections onto the nonlinear field pattern. The individual echoes are averaged to improve SNR.

The non-bijectivity of the approximately quadrupolar spatial encoding magnetic field (SEM) causes image aliasing that we disambiguate using parallel imaging as described by Schultz et al.[3]. We use an 8-channel receive array (Fig. 2a) with 8 cm circular loops for this purpose. The coil sensitivity map (Fig. 2b) is modeled using the Biot-Savart law and is different for each magnet rotation angle. The coil sensitivity profiles and field maps at each rotation angle are used to form the encoding matrix. We solve for 2D images (in the Y-Z plane) using the Algebraic Reconstruction Technique (ART) and Preconditioned Conjugate Gradients [2,4,5,6].

For 3D imaging, encoding is needed along the axis of the cylinder (X direction) as well. Transmit Array Spatial Encoding (TRASE) is a complementary, B1 imaging method [7,8], that can be used for 1D encoding in the X direction of the described scanner. To enable TRASE encoding in the inhomogeneous magnetic field, broadband WURST pulses were used [9]. The combination of the rotating SEM method for 2D encoding and the TRASE method for 3rd axis encoding, enables a 3D imaging method in the portable scanner [10].

Figure 3 shows experimental images of a “MIT/MGH” phantom (CuSO4-doped water, 1.7 cm thick, 13 cm dia.) acquired with 7 Rx coils, 32 averages of a 6 spin-echo train (TR = 550 ms, echo spacing = 8 ms) using 91 magnet rotations of 2o. Data is also acquired from a single field probe at each rotation to track field drift for use during reconstruction (Fig. 3b). The acquisition time of 66 minutes results from serially acquiring data from the 8 coils into a single channel on the console and would be reduced to 7.3 minutes with true parallel acquisition. Figure 3c shows a 1cm thick lemon slice imaged using 5 Rx coils, 128 echo train (TR = 4500 ms, echo spacing = 8 ms), and 181 magnet rotations of 1°. The total acquisition time was 93 minutes (15.5 minutes w/ parallel Rx).

Figure 4 shows experimental 3D imaging results using a phantom with 3 water-filled compartments spaced 2 cm apart in X. A 1D projection along X shows the water-filled compartments. Three slices are reconstructed from data at X = -2cm, 0cm, and 2cm using the corresponding 2D B0 field maps.

The nonlinear SEM of the Halbach magnet results in variable resolution over the FOV, including notable blurring in the center where the encoding field is shallow (Fig. 3c). If a sufficiently-strong linear term was added to the SEM, the spatially uniform encoding field region would not coincide with the axis of rotation. In this case, the “blind-spot” moves around the object resulting in less severe blurring [11] (an idea that guides the design of future magnets for this system). Field map errors from temperature drifts are significant but can be measured and included in the encoding matrix and mitigated in the reconstructed image (Fig. 3b).

Using an inhomogeneous rotating magnet for spatial encoding in lieu of gradient coils, we have constructed and demonstrated a lightweight scanner for 2D MR imaging with minimal power requirements. The 2D proof-of-concept images from this nearly head-sized imager show the ability of this encoding scheme to produce sufficient spatial resolution and sensitivity for the detection and characterization of many common neurological disorders such as hydrocephalus and traumatic space-occupying hemorrhages. Future work perfecting the calibration methods is likely to bring experimental image quality closer to the theoretical limit, but the resolution of the current system is sufficient for identifying gross pathologies.